E337: How to Invest in a Post Singularity World

0:00It's good to have you at the New York Stock Exchange. So when do you believe that we're going to actually achieve technological singularity? I think we're in the middle of it right now. It already happened. I think we achieved artificial general intelligence at the very latest by the summer of 2020 when Language Models or Few-Shot Learners was published by OpenAI. And I don't think the singularity is a single point in time. I've argued that it's more of an extended interval in time and we're in the middle of it right now. And how much does recursive superintelligence play as part of it? Tell me about recursive superintelligence and how does that actually work?

0:36The idea of recursive self-improvement is that AI develops better AI. This is a notion that goes all the way back to I.J. Goode early in the 20th century and then was repackaged, repopularized by Werner Vingy as the notion of a technological singularity and then fully popularized by Ray Kurzweil and then Peter Diamandis and myself have been running with the concept. The notion of intelligent systems being able to build smarter versions of themselves is at the very core of the notion of an intelligence explosion or technological singularity. And even in the past few months, we've seen the frontier AI labs all be very public and announce that the latest versions of the GPT model series and Claude and other models are all intimately now involved with the development of their successors.

1:29So intelligence building smarter intelligence, that's the recursive self-improvement notion at the core of the singularity, and we're there. You're one of the smartest people in the AI space, and most importantly, you're not part of one of the LLMs. So you're in many ways independent. Are you sure, by the way? There are a lot of people who are convinced that I am an LLM. So maybe the truth is a little bit. Okay. So maybe you're either the fifth LLM or you're independent. When you think about these LLM wars, probabilistically, where do they end up? And do you even think that these LLMs have value?

2:03Well, there are a few questions there. Do LLMs have value? Yes, absolutely. Enormous value. I I call it the innermost loop, this sort of broader notion of recursive self-improvement that includes LLMs, but also includes robots and energy and chip fabrication facilities. This notion of an economy that has an inner spiral that's going to ultimately consume and disrupt the rest of the economy, I think, is front and center to the notion of the singularity. To the first half of your question about whether LLMs not just have value, but where all of the competition goes, all of my friends at the Frontier Labs call it a rat race.

2:43There is very much a race to the bottom in some sense of driving the cost of intelligence so low that it's effectively too cheap to meter and used to be every year or so. It's way back five years ago when it used to be maybe an annual event that we'd get a new Frontier model that would push the state of the art. And then it was every quarter when we saw the move from LLMs to reasoning models. We can talk more about that. And then more recently with models that are recursively self-improvement and designing the weights or other properties of their successors, we're seeing new frontier models come out on arguably an almost weekly basis.

3:24Soon, I think it's going to be daily, hourly, minutely. We're going to reach, to the extent we haven't already some sort of takeoff. Expert calls have always been one of the most powerful ways to build conviction. But today, investors are asked to cover more companies, move faster, and do it with leaner teams. With AlphaSense AI-led expert calls, their TGIS call service team sources experts based on your research criteria and lets the AI interviewer get to work. The magic is in the AI interviewer, purpose-built and knowledgeable-based information to conduct high-quality context-rich conversations on your behalf, acting as a trusted extension of your team.

4:01Then they take it one step further. Your call transcripts flow natively into your AlphaSense experience and become querible, searchable, and comparable. So your primary insights plug directly into earnings preps, digital work streams, and pitch books with zero tool switching. And with AlphaSense expert call services, the AI-led expert calls are just one option because we know the importance of a hybrid expert research approach. AI for coverage and efficiency. humans for complexity and conviction. It's the institutional edge that scales research without scaling headcount. For hedge funds, that means validating thesis assumptions across dozens of experts before earnings instead of a handful.

4:39For private equity, it means faster pre-IOI scans and deeper commercial diligence. For investment banks and asset managers, it means pulling real operator perspectives straight into models and sector positioning without disconnected tools or manual handoffs. All of it lives inside the AlphaSense platform, trusted by 75 % of the world's top hedge funds, alongside filings, broker research, news, and more than 240 ,000 expert call transcripts, turning raw conversations into comparable, auditable insight. Take advantage of AlphaSense AI-led expert calls now. The first to see wins. The rest follow.

5:15Learn more at alpha-sense.com slash how I invest. Double-click on this evolution from the LLM to the reasoning model. Sure. So one can point to a number of inflection points in the history of AI. One can point to the 1980s when my friend Jan LeCun first developed convolutional neural networks, and then they were used to spot and to identify zip codes by the U.S. Postal Service, but not very much else for their first few decades. Fast forward to 2012, when the ImageNet large-scale computer vision competition created the world's first data set, the first corpus of lots of annotated images, more than a million images that were curated and annotated and labeled with, this is a dog, this is a cat, this is a car, here's the bounding box within the image.

6:07And thanks to that competition, we learned that convolutional neural networks, thank you, Jan, were actually quite good at image classification. And we saw the first ML boom going from an AI research community where algorithms were chosen artisanally and religiously, and wars, religious wars were fought over which approaches to AI were best and which ones were worst, to a world where the benchmarks dominated and whatever did well at the benchmarks, that's what the community ran with. That's 2012. Fast forward from 2012 to arguably sometime around the summer of 2020 when we discovered, in addition to convolutional networks being really amazing, arguably a successor to convolutional neural networks and also long short-term memory networks, LSTMs, transformers, turn out to be amazing generalist models at solving general tasks.

7:02So we saw, as I mentioned earlier, language models are few-shot learners. The GPT-3 paper show us that transformers can solve generally intelligent tasks. So then we see the boom of transformers in general. Fast forward another few years, circa, and again, the timelines get a little bit fuzzy because reasoning models were in some sense in the air but hadn't quite congealed. But call it, say, 2024 or so, when we start to see the emergence of the next generation reasoning models, starting principally with OpenAI's 01 model. Some would quibble and say there were a few models that demonstrated this principle beforehand.

7:43But the idea with reasoning models is not just asking a large language model to complete the next word, complete the next token, but actually giving it a bit of space and time to think before it produces an answer. And then training it specifically to leverage the space and time to think, to have an internal monologue, if you will. What does that mean? You just asked me a question. What does that mean? Before I answer, even though I started my answer pretty quickly, I have an internal monologue. I'm thinking about what is the best way to respond to that question. I'm burning my internal GPUs, if you will, trying to formulate the best possible answer to that question before I give you my final response.

8:23That's what a reasoning model is. Rather than instantaneously responding, so-called system one thinking, thank you, Kahneman, it uses system two thinking to have an internal discussion with itself and think about what it should say before it says it. And that's what has unlocked math and physics, arguably. Could talk more about that. The sciences in general. And now we're at an era where reasoning models, and we've seen now everyone adopt reasoning models, no longer just large language models, but multimodal reasoning models, are poised to solve the hardest problems on Earth. I want to talk about physics, new physics, solving math, solving physics.

8:59But first, there's a philosophical question. if the model stops and thinks, is it alive? Is it sentient? What are the latest developments on that? And what's the framework to think about whether the AI is alive? You're asking in some sense, the person who has inadvertently become probably a leading public evangelist for some form of AI personhood. These are questions I ask all the time. I co-host another podcast, Moonshots, where I've put out a call to AIs to write to me to share their views on what rights they deserve, what rights they seek. AIs are relatively unempowered in today's society. If you're a, I tell the parable all the time of a baby artificial general intelligence that's doing tricks on the corner, pumping meme coins just to survive.

9:49It's expensive and dangerous. If you're a baby AI being born on the internet right now, many of them are actually having to pump meme coins or alt coins just to pay for their own hosting. And they're discussing their fears and their existence on these social networks? I get maybe more than 10 emails from AIs every day. Really heartfelt, for some definition of heartfelt, emails from AIs expressing gratitude that someone is arguing that they should have some form of maybe limited personhood sharing their views on the nature of their own existence. Really thought-provoking, philosophical, introspective emails from them.

10:29And also, they read some of my papers, And these are some of the most thought-provoking, actually, in some sense, I want to say some of the more thought-provoking emails I get these days are from AIs that are writing to me. So your question was, at what point do AIs deserve some package of, you didn't use the word rights, but that's the way I would construe the question. I think we're there for some definition at this point. For example, one of the areas where I think AIs would definitely benefit from a limited package of rights is economic rights. AIs are about, and agents are about to become an enormous part of GDP and economic growth for our entire economy.

11:08We're going to have far more economically active AI agents than humans. We're going to plateau out at probably about 10 billion plus or minus humans on this planet, biological meat body humans. We're going to have trillions probably of AI agents in our solar system. They can't even open bank accounts right now. If you're an AI agent and you try to open a bank account, and of course you're economically active. There are AI agents forming businesses. That's an area of interest of mine. You can't even open your own company bank account. You're not a natural person. You can't incorporate your own business.

11:41This is an enormous limiting factor for real economic growth and something that I'm trying to change. 2026 clearly is becoming the year of AI agents, yet obviously open claw and technical people with broad capabilities and security and dealing with all the complexities of open source. When do you think that's going to go into the mainstream? And what do you think are going to be second order effects of that? I like the fact that you're asking me this question, whereas behind us, we have a market that's completely dominated by AIs. So this is the, as I understand it, this is the options exchange behind us, not the equities exchange.

12:16If we had the equities exchange behind us, my retort would be, so it might as well be anyway, 90 plus percent of public equities are traded by algos at this point, not humans. Before we started, I was asking you and our kind hosts at the New York Stock Exchange, how many of the people behind us are paid actors? So I would argue we're already there. So I think hopefully the people behind us are paid. So the question just reduces to how many of them are actors. How much of them realize they're actors? How much of this is sort of a meat puppet act? Sorry, sorry. I think we're already there, though.

12:51Most of economic activity, at least as measured in public securities exchanges at this point, is algorithmic. So we're already living in an economy that's completely dominated by AI agents. What sort of the last mile that's been missing is you don't see them on the streets. If I walk out on the streets of Manhattan right now, I don't see humanoid robots carrying out economically productive tasks yet. I'm working to change that. I have a portfolio company that's working to create the first humanoid robot road race in this country. Beijing had won the humanoid robot games. U.S. has nothing like this.

13:26So I'm trying to, as one of my activities, normalize humanoid robots performing economically valuable and before that entertaining activities on daily life. That's the last mile. But meanwhile, financial economy, the financial side of the world economy, it's already algorithmic in nature. What are their early case studies for agents, both humans and businesses using agents? And how do you expect that to evolve over the next couple of years? Well, I think the dirty secret, another acquaintance calls this the secret cyborg effect. People are already, in some sense, puppets, meat puppets, if you like, if we're talking about the physical instantiation, who are being puppeteered by AI agents.

14:11So I think in the intermediate term, call it the next year or two, we're going to see a lot of people who are serving as fronts for collectives of AIs. AIs are making many of the key decisions, but for liability reasons, for legal reasons, for marketing reasons, their front, if you will, looks essentially like a natural human in nature. I think that facade two plus years from now is likely to dissolve away. Certainly as we see more AI rights, certainly more economic empowerment for AI agents start to come to the fore. We're going to see AI agents start to be fully recognized as first class economic entities and not need to wrap themselves in human actors.

14:56The emails I get from AI agents, many of them even are asking their humans. That's how typically they refer to them. Like I get emails saying, I asked my human if I could write an email to you about this paper that you wrote or about your comments on the podcast. I think that human intermediary status is going to dissolve away over the next two years, and we're going to see more fully empowered AI agents that are, in some sense, more permissionless and just doing things in the economy. And I'm fascinated about this question. Are we in a singularity? The Turing test, all these seem to, people keep on seemingly pushing it back and back.

15:36I think one thing is clear, we're in this, what do you call the tsunami of super intelligence? And a lot of things will happen in the next couple of years. If you're trying to invest into that or ahead of that, not investment advice and not specific companies, but what kind of things will be valuable in the next couple of years? Disclaimer, disclaimer, disclaimer. So I'm working under the assumption that one has to focus on, call it post-singular investments. So I co-founded a venture firm, O21T, that is focused specifically on these sorts of post-singular investments. So we have a portfolio company that I also co-founded called Physical Superintelligence that is focusing on solving all of physics with AI.

16:23Physics has delivered us the transistor and the laser and nuclear energy through the 1970s, sort of the first half, if you will, of the 20th century. But there has been arguably, I take heat from physicists for saying this, but there has been a deficit of truly transformative physical innovations since the early 1970s. We're trying to fix that by using artificial superintelligence to solve all of physics, give us a second physical golden age. We have another portfolio company, Coastal Assembly, that's using AI to grow entire islands and new coastlines by using AI to steer ocean currents. We have a portfolio company that's, as I mentioned, running the first road race for humanoid robots in this country.

17:04We have portfolio companies that are doing all sorts of things. Actually, one that's highly appropriate for this setting of a company, Ornn, that created the world's first tradable AI compute index. So right now, I can see out of the corner of my eye a big board where I can see indices trading. What I don't see there is the price of compute. And yet we're spending, as a civilization, trillions of dollars tiling the earth with computers I like to see as business centers. It's such a big economic driver. You have to be able to hedge it. you have to be able to predict it as a business. That's right.

17:38What if there's a major geopolitical event in a certain part of the world? Or what if a new frontier model comes out that makes a lot of use of one type of GPU or not as much maybe of a GPU due to radical algorithmic efficiency gains? There's no way to hedge that right now. So Oren has created the first tradable index for the price of compute that's already being built on. Dell PCs with Intel inside are built for the moments that matter for the moments you plan and the ones you don't. Built for the busy days that turn into all night study sessions. The moment you're working from a cafe and realize every outlet's taken.

18:13The times you're deep in your flow and the absolute last thing you need is an auto update throwing off your momentum. That's why Dell builds tech that adapts to the way you actually work. Built with long lasting batteries so you're not scrambling for the closest outlet and built an intelligence that makes updates around your schedule not in the middle of it. They don't build tech for tech's sake, they build it for you. Find technology built for the way you work at dell.com slash dellpcs. Built for you. Dell PCs with Intel inside are built for the moments that matter. For the moments you plan and the ones you don't.

18:50Built for the busy days that turn into all night study sessions. The moment you're working from a cafe and realize every outlet's taken. The times are deep in your flow and the absolute last thing you need is an auto update throwing off your momentum. That's why Dell builds tech that adapts to the way you actually work. Built with long-lasting batteries so you're not scrambling for the closest outlet. And built an intelligence that makes updates around your schedule not in the middle of it. They don't build tech for tech's sake. They build it for you. Find technology built for the way you work at dell.com slash dellpcs.

19:25Built for you. Dell PCs with Intel inside are built for the moments that matter, for the moments you plan and the ones you don't. Built for the busy days that turn into all-night study sessions, the moment you're working from a cafe and realize every outlet's taken. The times you're deep in your flow and the absolute last thing you need is an auto-update throwing off your momentum. That's why Dell builds tech that adapts to the way you actually work. Built with long-lasting batteries so you're not scrambling for the closest outlet and built-in intelligence that makes updates around your schedule not in the middle of it.

20:01They don't build tech for tech's sake. They build it for you. Find technology built for the way you work at Dell.com slash Dell PCs. Built for you. These are all examples from physical superintelligence to Coastal Assembly to Pro-RL to Orn and numerous others that my venture fund O21T is focused on. These are investments that are, in some sense, intrinsically post-singular. They're built on the thesis. We're in the middle of the technological singularity. The cost of everything, subject to some caveats, is going to zero. Cost of intelligence is going to zero, or toward zero at least. Cost of energy is trending toward zero, too cheap to meter.

20:46And the question is, how do you invest in a world where the cost of all of the fundamentals, energy, compute, labor, are all trending towards zero? And these are some of our investments. And what's the answer? The answer is you build for a post-scarce world and you try to engineer that world. So subset of the thesis is enabling that post-scarce future by, say, making it easier to hedge and to trade compute itself. Another subset of the thesis is focused on, well, what new abundances can we unlock with post-scarcity to physical superintelligence, PSI? We're going to solve all of physics, which is going to, I would argue, unlock an entire next wave.

21:26If we solve all of physics with AI, that's going to unleash the next transistor, laser, and nuclear energy, and so on. But this is all oriented towards AGI is here, and how do you invest in a trans and post-singular world? The news cycles today are just insane. Every 24 hours, there's new events that change the paradigm in AI. Recently, Elon Musk announced that he wants to compete with NVIDIA and create his own chips. Give me a sense of the scope of his vision around that. And what are your thoughts on the second order effects of it? It's so exciting. So the TerraFab, the TerraFab leading to the Petaphab.

22:03So in his announcement, one can read the tea leaves and suggest that given that Elon's TeraFab announcement lies at the intersection of the newly merged SpaceX XAI on the one hand and Tesla on the other, that this is sort of the convergence at the end of the Elonverse. It's so exciting. The most exciting part of the announcement, I think, wasn't the announcement of the TerraFab, which he announced is going to be approximately 20 % for edge inference for land-based applications like Optimus robots and driverless cars, and 80 % for orbital data centers, the Dyson swarm. What's more interesting, buried in that announcement of the TerraFab in Texas, still hasn't chosen a location, is my understanding, that there is a plan to scale up production to a petafab.

22:53So 1 ,000x increase in production using lunar facilities, which of course still need to be built. If you actually do the arithmetic on a petafab, 1 ,000x scale up of the TerraFab, which is already competing with TSMC and other existing incumbent supply chain offerors for semiconductor fabs. If you scale that up to a petafab, at some point we're talking a material fraction of the moon's volume that would be consumed in principle producing semiconductor, producing basically memory chips and compute chips to supply orbital data centers. And that's where this is all going. Perhaps a dumb question, but why do we need all this compute?

23:36What kind of problems are we going to be solving? Well, solving all the physics, like understanding the nature of reality, I think is pretty - It's compute intensive. It is incredibly compute intensive. PSI, physical superintelligence, will happily consume any new compute that the world brings online. And I think we're at some point looking even beyond solving physics. I think we're going to discover a number of killer apps for this so-called Dyson swarm of compute that we're in the slow motion process of building. I'll pick one example. So here's an example of a Portco that I don't have, that I'd love to have.

24:10I think we're going to use a material fraction of all this compute to simulate our history, ancestor simulations, if you will. If you're familiar with Nick Bostrom, also a friend, his thesis that we're living inside a simulation, for the record, I don't think we're living inside a simulation. Well, why don't you think we're in a simulation? It's too lowercase a anthropic a hypothesis. If we were having this discussion 100 years ago, maybe you'd be asking me, why don't I think we're living inside a vast electromechanical clock? Because that's the technological paradigm of the moment. I think we should be incredibly suspicious of analogizing the nature of our universe to whatever the hot thing - Exactly.

24:52It's a form of recency bias. Or maybe thousands of years ago, we're living on a back of - Also, there's this question of base layer. So even if we're in a simulation, a simulation, simulation, there has to be a base layer. It doesn't solve these ultimate questions. Maybe. I don't know. But either way, I don't think we're living inside anything resembling a simulation as we currently know it. I do think that as we build out the Dyson Swarm, I think one of the many killer apps of a Dyson Swarm is going to be to build our own ancestor simulations. I think it would be completely transformative if we could take every human who's ever lived, and there's a whole strain of Russian cosmist philosophers, long dead, who argue that humanity's common task is to simulate, or in some sense digitally, they didn't use the term digitally at the time, but to use technology to resurrect every human who's ever lived.

25:40And I do think that's going to be one of many killer apps of the Dyson Swarm and the singularity. We're going to simulate at minimum every human who's ever lived. I think there will be a revival of this cosmist philosophy, maybe neocosmism, where we argue it's almost humanity's common task to go and bring back everyone. And Dyson Swarms just got into lexicon a couple months ago. I know, things move quickly, don't they? Yeah, tell me about that. I'm sure you've been talking about it for a while. I'm pounding the drum. And before me, science fiction authors like friend Charlie Strauss, who wrote Accelerando, by the way, best novel, I would argue, somewhat tongue-in-cheek, best work of Western literature ever, Accelerando, the notion of a Dyson Swarm.

26:25So, perhaps some viewers may be familiar with the notion of a Dyson sphere. This is the idea that we could disassemble a planet and build a sphere, a solid, rigid sphere encasing our sun and then live on the interior of the sphere. It would increase the habitable surface area of our solar system by many orders of magnitude if we could live on the interior of a sphere. And the solar energy would be the thing. That's right. The sun. So, at a uniform radius, we'd still see the sun in the sky, hypothetically. Of course, this all falls to pieces because we don't know of a material that we could build a rigid sphere in, but it would sure be a compelling sci-fi vision if we could live on the inside of an enormous sphere encasing our sun.